`read_sample` gives `UnexpectedEof` error in 0.9

[probablykasper]

Just tried updating to 0.9, but whenever I call read_sample I'm getting this error:

Err(IoError(Error { kind: UnexpectedEof, message: "failed to fill whole buffer" }))

The mpeg_aac_decoder example seems to work fine, so there might be something I could do differently, but still this did work fine in 0.8.

Here's the branch where I tried the upgrade: https://github.com/probablykasper/redlux/tree/mp4-0.9
commit: probablykasper/redlux@cc8add1

[NightlySide]

It seems that the piece of code raising the error is the following one:

mp4-rust/src/track.rs

Lines 526 to 535 in 56cc260

	let sample_offset = match self.sample_offset(sample_id) {
	Ok(offset) => offset,
	Err(Error::EntryInStblNotFound(_, _, _)) => return Ok(None),
	Err(err) => return Err(err),
	};
	let sample_size = self.sample_size(sample_id).unwrap();
	let mut buffer = vec![0x0u8; sample_size as usize];
	reader.seek(SeekFrom::Start(sample_offset))?;
	reader.read_exact(&mut buffer)?;

[alfg]

Thanks for catching this. Will take a look as soon as I can.

[Serial-ATA]

@alfg I think I found the issue

mp4-rust/src/mp4box/tfhd.rs

Line 61 in 93dbb56

let base_data_offset = reader.read_u64::<BigEndian>()?;

The base data offset is always assumed to be present, but I checked how FFmpeg handles this box:

https://github.com/FFmpeg/FFmpeg/blob/2b2380c098569af88e7a77aff78f330d11090512/libavformat/mov.c#L4819-L4821

The base data offset is dependent on the flags. I checked the asset in probablykasper/redlux#4, which had an implicit offset, so the offset was actually being read from the stsd id and default duration, making it incredibly large (4294968320, greater than the file size).

I can make a PR to fix this, but I'm not sure how you would want this done. Since you're using the ReadBox impl I can't pass along the moof offset.

[alfg]

Hey @Serial-ATA, thanks for checking into this. Sorry, I haven't had much time lately to work on this, but I'm open to suggestions and PRs.

[Serial-ATA]

I'll have to either remove the ReadBox impl or somehow return the flags from it, maybe by storing them in the TfhdBox itself? (Can't do that since the box could theoretically be empty besides the version, flags, and id)

That'll break consistency. What do you think?

[probablykasper]

@Serial-ATA any clue why it worked before 0.9?

[Serial-ATA]

Could be #46? Not sure. A quick look through the history didn't turn up anything obvious, and I don't really know the project. 😄

[w-flo]

I suspect this might be fixed in git master thanks to #86 and #106. I don't have a file that is affected – maybe someone in here could test this?

[probablykasper]

It seems to be fixed, yeah!